City, University of London Institutional Repository

Citation: Velasco, C., Woods, A. T., Marks, L. E., Cheok, A. D. ORCID: 0000-0001-6316-2339 and Spence, C. (2016). The semantic basis of taste-shape associations. PeerJ, 4(e1644), doi: 10.7717/peerj.1644

This is the draft version of the paper.

This version of the publication may differ from the final published version.

Permanent repository link: http://openaccess.city.ac.uk/19502/

Link to published version: http://dx.doi.org/10.7717/peerj.1644

Copyright and reuse: City Research Online aims to make research outputs of City, University of London available to a wider audience. Copyright and Moral Rights remain with the author(s) and/or copyright holders. URLs from City Research Online may be freely distributed and linked to.

City Research Online: http://openaccess.city.ac.uk/ publications@city.ac.uk

City Research Online

Submitted 11 September 2015Accepted 10 January 2016Published 4 February 2016

Corresponding authorCarlos Velasco,carlos@imagineeringinstitute.org

Academic editorTifei Yuan

Additional Information andDeclarations can be found onpage 15

DOI 10.7717/peerj.1644

Copyright2016 Velasco et al.

Distributed underCreative Commons CC-BY 4.0

OPEN ACCESS

The semantic basis of taste-shapeassociationsCarlos Velasco1,2, Andy T. Woods1,3, Lawrence E. Marks4,5, Adrian David Cheok2,6

and Charles Spence1

1Crossmodal Research Laboratory, Department of Experimental Psychology, University of Oxford,Oxford, UK

2 Imagineering Institute, Iskandar, Malaysia3Xperiment, UK4 Sensory Information Processing, John B. Pierce Laboratory, New Haven, CT, USA5 School of Public Health and Department of Psychology, Yale University, New Haven, CT, USA6 School of Mathematics, Engineering, and Computer Science, City University, London, UK

ABSTRACTPrevious research shows that people systematically match tastes with shapes. Here, weassess the extent to which matched taste and shape stimuli share a common semanticspace and whether semantically congruent versus incongruent taste/shape associationscan influence the speed with which people respond to both shapes and taste words. InExperiment 1, semantic differentiation was used to assess the semantic space of bothtaste words and shapes. The results suggest a common semantic space containing twoprincipal components (seemingly, intensity and hedonics) and two principal clusters,one including round shapes and the taste word ‘‘sweet,’’ and the other including angularshapes and the taste words ‘‘salty,’’ ‘‘sour,’’ and ‘‘bitter.’’ The former cluster appearsmore positively-valencedwhilst less potent than the latter. In Experiment 2, two speededclassification tasks assessed whether congruent versus incongruent mappings of stimuliand responses (e.g., sweet with round versus sweet with angular) would influence thespeed of participants’ responding, to both shapes and taste words. The results revealedan overall effect of congruence with congruent trials yielding faster responses thantheir incongruent counterparts. These results are consistent with previous evidencesuggesting a close relation (or crossmodal correspondence) between tastes and shapecurvature that may derive from common semantic coding, perhaps along the intensityand hedonic dimensions.

Subjects Psychiatry and PsychologyKeywords Semantic differentiation, Shapes, Crossmodal correspondences, Tastes,Congruency effects

‘‘She laughed, a laugh sweeter than honey, with a sound curving and zigzagging, as ifsinging’’

(Mo Yan, Ball-Shaped Lightning, cited by Yu, 2003, p. 190)

INTRODUCTIONSeveral studies show that people systematically match both basic taste words and tastants(i.e., chemicals that generate gustatory sensations) with shapes that vary in terms of

How to cite this article Velasco et al. (2016), The semantic basis of taste-shape associations. PeerJ 4:e1644; DOI 10.7717/peerj.1644

1Here, we are mainly interested in the wayin which taste/shape correspondencesoccur in the general population. As aconsequence, research on synaesthesia,a rare condition where the stimulation ofone sensory modality leads to concurrentsensory experiences in the same orother modalities, falls out of the scopeof the present study. It is important tomention, though, that cases of taste-shapesynaesthesia have been reported elsewhere(Cytowic, 1993; Cytowic & Wood, 1982).

their curvature (see Cytowic & Wood, 1982; Spence & Deroy, 2014; Spence & Ngo, 2012,for reviews). Over the last few years, researchers, including ourselves, have studiedcrossmodal (taste-shape) correspondences,1 providing some hints as to their underlyingmechanisms (e.g., Velasco et al., 2015a; Velasco et al., 2016) and their effects on taste (orgustatory) information processing more generally (e.g., Gal, Wheeler & Shiv, 2007; Lianget al., 2013). Notably, while the way in which people match basic tastes with shapes seemsreasonably well understood, the mechanisms that underlie crossmodal correspondences,as revealed in crossmodal (taste-shape) matches and congruency effects in perceptualprocessing are still to be clarified. In particular, research still needs to clarify when, how,and why the mechanism(s) that underlies taste-shape correspondences may influence theprocessing of taste (perceptual and linguistic) and shape information.

Velasco and his colleagues have investigated how people match basic tastes and shapes.So, for example, the results of one series of four experiments revealed that people associatesweet (both when presented as a word and as a tastant) with round shapes, and bitter,salty, and sour (as words and tastants) with more angular shapes (Velasco et al., 2015a,see also Ngo et al., 2013; Velasco et al., 2014; Velasco et al., 2015b;Wan et al., 2014). Whatis more, Velasco et al. (2015a) also reported that the more the participants liked the taste(but not a taste word), the rounder the shape matched to it (see also Bar & Neta, 2006, oncurved objects preference) and suggested a hedonic mechanism to explain the crossmodalmatching (see also Ghoshal, Boatwright & Malika, 2015). This finding was subsequentlyreplicated by Velasco et al. (2016). The latter researchers found that taste concentrationcan also affect shape matching, with more versus less intense tastants more likely matchedto angular versus round shapes, respectively (see also Becker et al., 2011). Given the focusof the present study—on the semantic basis of taste word/shape correspondences, andassociated congruence effects—the aforementioned findings are intriguing. Nevertheless,correlations alone do not suffice to show that a hedonic mechanism underpins thecorrespondences. Moreover, it is important to evaluate a wider range of intensities, giventhat the authors tested just two concentrations.

The idea that certain crossmodal correspondences may be mediated by the affectiveproperties of the matching or mismatching stimuli is certainly not new (e.g., Kenneth,1923; see alsoMarks, 1996, for a review). Indeed, researchers have demonstrated recentlythat the way in which people match music and colours (Palmer, Langlois & Schloss,2015; Palmer et al., 2013) and colours and fragrances (Guerdoux, Trouillet & Brouillet,2014; Schifferstein & Tanudjaja, 2004) can be mediated by emotion. In an earlier study,Collier (1996) demonstrated that judgments of many sets of visual and auditory stimulicould be reduced to two dimensions, identified, by the author, as valence and activity.Further, some of the visual and auditory stimuli overlapped on these dimensions.Collier’s generalization is relevant to taste/shape correspondences, which also mightoverlap on the same affective dimensions. What is more, such an idea is important inthe context of multisensory perception in general, given that, in addition to factorssuch as spatiotemporal alignment and semantic congruence (Stein, 2012), crossmodalcorrespondences can also help mediate multisensory perception (e.g., Parise, Knorre &

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 2/20

Ernst, 2014). Indeed, crossmodal correspondences can provide relevant information topeople when they make inferences about the (often) noisy sensory world (Parise, 2015).

Importantly, earlier research also points to the notion that taste/shape correspondencesmay influence the processing of taste-information. For instance, Liang et al. (2013)assessed the influence of shapes on people’s sensitivity to sweetness using near-thresholdsucrose solutions. In their study, people rated round shapes as more pleasant. Further,presenting a round shape rather than an angular shape before tasting a sweet solutionenhanced sweetness sensitivity. Unfortunately, however, this study is the only of itskind, and replication is critical (the effect is certainly specific and small), perhaps usingeveryday, suprathreshold, solutions. Moreover, there is a possible confound of responsebias in the study, as Liang et al. (2013) did not attempt to control for the participants’response criterion (e.g., apparently they did not include any ‘blank,’ water trials).

Gal, Wheeler & Shiv (2007), reported a study in which their participants were askedto indicate which of three shapes (which could be all round or angular) had the largestsurface area before rating a piece of cheddar cheese. The results showed that the curvatureof the shapes presented in the first task influenced the perceived sharpness of the cheese,with the angular shapes leading to higher sharpness ratings than the round shapes. Whatis more, other studies have shown that the shape of a plate and food (when it is round ascompared to angular) can influence participants’ sweetness ratings of the food (resultingin people rating the food as tasting sweeter, see Fairhurst et al., 2015; Stewart & Goss, 2013;see also Piqueras-Fiszman et al., 2012).

Here it is worth mentioning that, in spite of their perceptual basis, similarities acrossthe senses also surface in language (e.g., see the quote at the beginning of the Intro-duction,Marks, 1978;Marks, 1996). With this in mind, we ask whether the potentialhedonic- and intensity-related explanations/mediations of taste/shape correspondencesmay extend to taste words and, if so, whether they reflect: a perceptual process; a commonconnotative meaning (Walker, 2012;Walker, Walker & Francis, 2013; see also Collier, 1996;Karwoski, Odbert & Osgood, 1942, for earlier examples); or perhaps, a combination of thetwo (see alsoWalker & Walker, in press). According to the semantic coding hypothesis(SCH, seeMartino & Marks, 2001), high level mechanisms that connect informationacross the senses may emerge from developmental experiences with various perceptsthat are coded into language, and that can affect multiple levels of human informationprocessing. Consequently, crossmodal congruence effects can arise not only in theprocessing of perceptual stimuli, but also in the processing of verbal stimuli (Martino &Marks, 1999). While Liang et al. (2013) provided some evidence that congruent shapes caninfluence people’s detection of sweet solutions presented at near threshold levels, we askhere whether the congruence of taste words and shapes can affect perceptual processing.AsMarks (1978) pointed out, ‘‘According to the Oxford English Dictionary, ‘sharp’applied first to touch, then subsequently to taste (ca. 1,000), visual shape (1,340), andhearing’’ (p. 190), indicating that shape-related words have been used to describe tastesfor several centuries, and thereby perhaps some kind of implicit relation between shapeand taste quality (see alsoWilliams, 1976; Yu, 2003).

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 3/20

Here, we describe two experiments designed to assess whether shapes and taste wordsshare a common semantic space and whether congruence between them can influenceboth taste words and shape information processing. Experiment 1 used semanticdifferentiation (Osgood, Suci & Tannenbaum, 1957) to assess whether taste words andshapes share common dimensions of connotative meaning. Experiment 2 used a speededclassification task to assess whether taste/shape congruence affects the categorization oftaste words and shapes. We hypothesized that taste words and shapes share a commonsemantic space to which previously reported associations will project, and that people willrespond faster to the congruent versus incongruent pairings.

EXPERIMENT 1Methods and materialsParticipantsA total of 102 participants (M age= 34.7 years, SD= 11.8, age range= 19–70, 51females) took part in the study, online through the Adobe Flash based Xperimentsoftware (http://www.xperiment.mobi). The participants were recruited using Amazon’sMechanical Turk in exchange for a payment of 1.50 USD (seeWoods et al., 2015, for amethodological overview of internet-based research). All of the participants were basedin the USA, and all agreed to take part in the study after reading a standard consent form.The experiment was reviewed and approved by the Central University Research EthicsCommittee at the University of Oxford (MS-IDREC-C1-2014-056).

Apparatus and materialsThe images of four shapes (previously used by Köhler, 1929; Ramachandran & Hubbard,2001), two angular and two round (see Fig. 1), as well as four taste words, namely bitter,sour, salty, and sweet, were the stimuli in this study. The taste words were presented infont Times New Roman 80.

Each stimulus was assessed using the semantic differential technique (SDT). This is aprocedure in which the connotative meaning of objects and concepts is measured by usingrating scales with polar adjectives (see the original work of Osgood, Suci & Tannenbaum,1957, for details; see also Albertazzi et al., 2014, for a recent example). Twelve pairsof polar adjectives were included, which were based on previous research using theSDT (Osgood, Suci & Tannenbaum, 1957; Osgood, 1964). Each pair has been found tocorrelate with three bipolar dimensions, namely, evaluation, potency, and activity. Thepairs of adjectives were: (1) nice–awful, (2) good–bad, (3) mild–harsh, (4) happy–sad(evaluation), (5) powerless–powerful, (6) weak–strong, (7) light–heavy, (8) shallow–deep (potency), (9) slow–fast, (10) quiet–noisy, (11) passive–active, and (12) dead–alive (activity). Each shape and taste stimulus was rated on a 100-point visual analoguescale (VAS), unmarked except for the adjectives, located outside the poles of the scale.Adjectives within the pairs 1, 3, 6, 8, 10, and 12 were reversed during in the experiment.

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 4/20

Figure 1 Shape stimuli used in Experiment 1.

Figure 2 Example of (A) A taste word and (B) A shape trial in Experiment 1.

ProcedureAt the beginning of the study, all participants were informed about the general aimsand agreed to take part after reading a standard consent form. In the instructions, theparticipants were told that they would be presented with taste words or shapes and askedto rate them on a number of different VAS scales. On each trial, one of the stimuli (shapeor taste word) was presented in the middle of the screen together with a VAS (see theexample in Fig. 2). Trials were blocked by pair of adjectives (scales), and both order oftrials and order of blocks were randomized across participants. In each block of adjective-defined scales, participants responded to the eight stimuli (four shapes and four tastewords), giving rise to a total of 96 trials.

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 5/20

Table 1 Varimax-rotated component matrix in Experiment 1 (see also Fig. 4).

Adjectives Component

1 2

Passive–active .995 −.018Slow–fast .986 .033Powerless–powerful .986 −.065Weak–strong .980 −.125Shallow–deep .934 −.182Quiet–noisy .933 −.184Harsh–mild −.825 .565Sad–happy .168 .979Awful–nice −.293 .940Bad–good −.311 .919Light–heavy .497 −.825Dead–alive .560 .808Eigenvalues 7.09 4.43% of variance 59.09% 36.89%

AnalysisA varimax-rotated principal component analysis (PCA) was used to define the principaldimensions arising from the different scale ratings of tastes and shapes. In addition, ahierarchical cluster analysis with Ward’s method and squared Euclidean distance as thesimilarity measure (see Kaufman & Rousseeuw (2005), for more details) was conductedin order to assess whether the different tastes and shapes would group as a functionof common ratings in the scales used in Experiment 1. The aforesaid analyses wereperformed with IBM SPSS v. 19 and the PCA visualizations were created in the R′ (RCore Team, 2015) {FactoMineR} package (see http://factominer.free.fr/). The data wereaggregated as a function of dimensions and clusters and Wilcoxon signed-rank tests wereperformed in R, to assess any difference between clusters as a function of dimensions.Effect sizes were calculated by means of Cliff’s Delta as implemented in the {effsize}package in R (see https://cran.r-project.org/web/packages/effsize/effsize.pdf), in which0 indicates the absence of an effect (the distributions overlap), while a value of−1 or 1indicates a large effect (no overlap whatsoever; see Cliff, 1996).

Results and discussionThe principal component analysis (PCA, see Fig. 3) revealed that two components hadeigenvalues over Kaiser’s criterion of 1 and, in combination, explained 95.98% of thevariance. Table 1 shows the factor loadings after the varimax (orthogonal) rotation. Notethat the first and second components accounted for 59.09% and 36.89% of the variance,respectively.

The dendrogram resulting from the hierarchical cluster analysis appears in Fig. 4. Twomajor clusters are evident (see Table 2), one grouping round shapes with the taste word‘‘sweet,’’ and another grouping angular shapes with the taste words ‘‘salty,’’ ‘‘sour,’’ and

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 6/20

Figure 3 (A) Unrotated factor map of the polar scales in Experiment 1. Note that only the label of theupper end of the scales is presented. (B) Unrotated factor map for the stimuli. The circles grouped thevariables as a function of the two clusters identified in the subsequent cluster analysis. Note that giventhat (A) and (B) show the unrotated visualizations, the percentages for each component vary slightly fromthose presented in Table 1.

‘‘bitter.’’ These groupings reflect the tendency for stimuli in each cluster to receive similarratings on the different semantic differential scales.

After identifying the two principal components and the two clusters, the data wereaggregated as a function of dimension and cluster (Fig. 5 summarizes the mean values).Note that the scores of harsh/mild and light/heavy were reversed as they correlatednegatively with their respective dimensions. Wilcoxon signed-rank tests were performedin order to assess any difference between clusters on each dimension. The ratings on the

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 7/20

Table 2 Hierarchical cluster analysis in Experiment 1.Note that the distance measure was rescaled to a∼0–1 range, and that the first large jump occurs between stages four and five.

Stage Cluster combined Coefficients Stage cluster first appears Next stage

Cluster 1 Cluster 2 Cluster 1 Cluster 2

1 Sour Bitter .000 0 0 42 .010 0 0 6

3 .031 0 0 5

4 Salty Sour .072 0 1 65 Sweet .220 3 0 7

6 Salty .400 2 4 7

7 1.155 6 5 0

first dimension of the stimuli in the second cluster were higher than those in the firstcluster (p < .001, Cliff’s Delta= 0.96), whereas the ratings on the second dimension ofthe stimuli in the first cluster were lower than those in the second cluster (p< .001, Cliff’sDelta= 0.79). In other words, the round shapes and the taste word ‘‘sweet’’ were ratedas more positively-valenced and less intense than the angular shapes and the taste words‘‘salty,’’ ‘‘sour,’’ and ‘‘bitter.’’

These results provide further support for the presence of an association between theword ‘‘sweet’’ and round shapes and the words ‘‘bitter,’’ ‘‘salty,’’ and ‘‘sour’’ and angularshapes (Velasco et al., 2015a; Velasco et al., 2015b; Velasco et al., 2016). Moreover, theresults also suggest that tastes and shapes share a semantic space, or a set of implicitmeanings, which is initially characterized by two main components. Indeed, a possibilityis that these components reflect the two elements identified by Velasco et al. (2015a) andVelasco et al. (2016), namely hedonic value and intensity. Consistently, the results ofExperiment 1 fall in line with the idea that perceptual dimensions (e.g., sweet vs. sour,and fast vs. slow) differentiate between valence and arousal in specific ways (e.g., positiveand negative and high and low arousal, respectively, see Cavanaugh, MacInnis & Weiss,2015). Two limitations may be mentioned in regard to this study. First, it is possiblethat blocking by pairs of adjectives may have increased contrast between stimuli in eachdimension. Second, the number of shapes included is very limited. This said, the commonsemantic space between round shapes and angular shapes, as well as their differences, arecertainly consistent with previous research using those shapes (Holland & Wertheimer,1964; Lindauer, 1990; Lyman, 1979).

Experiment 1 provides evidence in support of the idea that taste words and visualshapes share dimensions of connotative meaning. Given this, Experiment 2 aimed toassess whether the crossmodal correspondence between taste words and shapes wouldproduce congruence effects over-and-above those already reported with tastes per se

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 8/20

Figure 4 Dendrogram obtained by means of hierarchical cluster analysis in Experiment 1.

Figure 5 Mean ratings for each cluster and dimension in Experiment 1. The error bars represent thestandard error of the means.

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 9/20

Figure 6 (A) The shape stimuli used in both tasks, (B) A trial in the shape response task, and (C) A trialin the taste response task.Note that the shape stimuli are group in pairs as used as responses for the shaperesponse task.

(Liang et al., 2013), that is, by using linguistic taste stimuli. For this purpose, a task wasdesigned in which a larger sample of participants (in order to compensate for potentialhardware-related differences across participants and fewer trials, e.g.,Woods et al., 2015)were given congruent or incongruent instructions about the mapping between taste wordsand shapes and were later asked to respond to shapes or taste words with taste words andshapes, respectively.

EXPERIMENT 2Methods and materialsParticipantsA total of 253 participants (M age= 34.48 years, SD= 10.90, age range= 18–73 years,138 females) took part in the study online and received a payment of 1.80 USD. All werebased in the USA, and all agreed to take part in the study after reading a standard consentform.

Apparatus and materialsThe ten stimuli comprised four pairs of shapes (one round and one angular within eachpair, 200× 200 pixels each; see Fig. 6A), plus two taste words ‘‘sweet’’ and ‘‘sour.’’ Thetaste words were again presented in font Times New Roman 80.

ProcedureThe participants took part in two tasks. In one of the tasks (shape response, see Fig. 6A),the participants were presented the taste words ‘‘sweet’’ or ‘‘sour’’ (one at a time) andasked to respond with either an angular or round shape (i.e., pairs of shapes taken fromExperiment 1, see Fig. 6B). In the other task (taste response, see Fig. 6C), the participantswere presented the eight shape stimuli (one at a time) and were asked to respond with thetaste words ‘‘sweet’’ or ‘‘sour.’’ Taste/shape congruence was manipulated in both tasks.

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 10/20

Table 3 Experimental design used in Experiment 2.

Task Congruence Instructions (stimulimapping)

Stimuli Responses Uniquetrials

Repetitions

Congruent Sweet–round andsour–angular

8Shaperesponse Incongruent Sweet–angular and

sour–round

Taste words(sweet or sour)

Angular orround shape(four pairs×2)

8

Congruent Round–sweet andangular–sour

8Tasteresponse Incongruent Round–sour and

angular–sweet

Shapes (eightshapes) Sweet or sour

8

X2

That is, each task included a block of congruent trials and a block of incongruent trials.Both task order and block order were randomized across participants. In the congruent(incongruent) block of the taste response task, the participants were asked to respond withthe word sweet every time they saw a round (angular) shape and with the word sour everytime they saw an angular (round) shape. In the congruent (incongruent) block of theshape response task, the participants were instructed to respond with round shapes everytime they saw the word sweet (sour), and with angular shapes every time they saw theword sour (sweet). Note, however, that the possible responses were presented to the leftor to the right of the target stimulus, and the participants would have to press z or m, afunction of the position of the correct response (see below). In both tasks, the participantswere instructed to respond as rapidly and accurately as possible to a target stimulus (tastewords or shapes) by pressing the key that was associated the stimulus that matched theparings in the instruction (shapes and taste words, respectively).

Table 3 summarizes the experimental design. Each of the tasks included eight uniquetrials. In the shape response task, half of the trials required of the participants to respondto the word ‘‘sweet’’ and the other half to the word ‘‘sour.’’ Moreover, four trials includedthe round shapes on the right and the angular on the left (two for ‘‘sweet’’ and two for‘‘sour’’); in the remaining four trials, the positioning was reversed. In the taste responsetask, the participants responded to the eight shapes with words ‘‘sweet’’ and ‘‘sour.’’ Theright-left position of ‘‘sweet’’ and ‘‘sour’’ was thus fully counterbalanced.

All eight unique trials were presented, once each for practice, before each block ofcongruent and incongruent trials in each task. Feedback came after each of the practicetrials with the word ‘‘correct’’ or ‘‘wrong’’ presented for 0.5 s. Immediately after thepractice trials, the participants proceeded to the experimental trials of the block. All eightunique trials were presented twice, giving rise to a total of 16 trials per block and 64 forthe whole experiment. To prevent the participants from responding simply to the right orleft position as opposed to sweet/sour or angular/round, 8 trials in each block mapped toone response (‘‘sweet’’/‘‘sour,’’ angular/round) to the z key, and in the other 8 trials to them key.

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 11/20

AnalysesBoth accuracy and RTs were analysed as a function of task and congruence. Accuracyand RTs were analysed by means of 2× 2 analysis of variance-type statistics (ATS) withthe factors of task and congruence. Note that the ANOVA-type statistic, a robust rank-based nonparametric alternative to the classic ANOVA, is robust to both outliers and theviolation of assumptions in classical parametric ANOVA (see Erceg-Hurn & Mirosevich,2008). The analyses were performed in R Statistical Software, as implemented in the{nparLD} package (Noguchi et al., 2012). The significant main effects and interactionswere further analysed with the Wilcoxon signed-rank test to which Bonferroni correctionswere also applied. Effect sizes were also calculated by means of Cliff’s Delta.

Results and discussionAccuracyData from those participants failing to respond accurately on more than 60% of thetrials were excluded from the analyses (a total of 14 participants). Whilst there was asignificant main effect of task, FATS(1,∞)= 15.15, p< .001, the effect of congruence wasnot significant, FATS(1,∞)= 0.39, p= .530, nor was the interaction between task andcongruence, FATS(1,∞)= 1.68, p= .195. Wilcoxon signed-rank test revealed that theparticipants were more accurate in the task in which they had to respond with taste wordsrather than shapes (p= .001, Cliff’s Delta= 0.15). Figure 7A summarizes the results.

RTsOnly the trials in which the participants responded correctly were included in the analyses(91.6% of 15,296 trials). The ANOVA-type statistic revealed a significant effect of task,FATS(1,∞) = 121.31, p < .001, and congruence, FATS(1,∞) = 13.71, p < .001. Theinteraction between task and congruence was not significant, FATS(1,∞)= 3.38, p=0.066. The participants responded more rapidly in the task in which they respondedwith taste words rather than with shapes (p < .001, Cliff’s Delta= 0.27). Moreover,participants also responded more rapidly on the congruent than the incongruent trials(p< .001, Cliff’s Delta= 0.13).

The results of Experiment 2 provide evidence for the idea that taste/shape correspon-dences can indeed produce congruence effects even in the absence of a tastant, but justwith shapes and taste words. It is worth mentioning, though, that there was a differenceacross tasks too: That is, the participants responded more accurately, and more rapidly,in the taste response task as compared to the shape response task. Moreover, an overallcongruence effect was observed across tasks. The results of Experiment 2 extend previousstudies assessing taste/shape congruence (e.g., Fairhurst et al., 2015; Liang et al., 2013) totaste word/shape congruence.

GENERAL DISCUSSIONTwo experiments aimed to assess, first, whether basic taste words and shapes share asemantic space—a set of implicit meanings—that may contribute to the correspondencesbetween these stimuli, and, second, whether these crossmodal correspondences can induce

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 12/20

Figure 7 Summary of the results of Experiment 2. (A) Accuracy and (B) Mean reaction times (RTs) inboth tasks as a function of congruence. The error bars represent the standard error of the means.

congruence effects when linguistic stimuli rather than tastants are used. Experiment1 revealed that taste words and shapes can share a semantic space, which is mainlycharacterized by dimensions related to intensity and hedonic value. Moreover, consistentwith previous research (Spence & Deroy, 2014, for a review), specific tastes and shapesclustered, ‘‘sweet’’ with round shape and ‘‘bitter,’’ ‘‘salty,’’ and ‘‘sour’’ with angularshape. A limitation of Experiment 1, though, is that only four shapes were used.Nevertheless, the results are consistent with previous research suggesting that theshapes used are distinctively round and angular (e.g., Holland & Wertheimer, 1964;Lindauer, 1990; Lyman, 1979), that the aforesaid groups of tastes and shapes tend to be ratedsimilarly (Velasco et al., 2015a; Velasco et al., 2016), and that perceptual dimensions can bedescribed along such dimensions (Collier, 1996; Cavanaugh, MacInnis & Weiss, 2015).

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 13/20

2This is of particular relevance giventhe fact that previous research hasdocumented the importance of somenon-semantic features of taste words(e.g., typeface features, see Velasco etal., 2015b) and/or implicit vocalization,articulation/kinesthesis, and/or soundimagery in conveying meaning (e.g., Ngoet al., 2013). For example, one may arguethat it is not the word ‘‘sweet’’ but ratherits sound symbolic meaning, that guidesits matching to round shapes. While wecannot rule out all the specific interactionsbetween the aforesaid variables, it is knownthat tastants and taste words are similarlymatched to shapes varying in terms of theircurvature (Velasco et al., 2015a).

Experiment 2 introduced a task in which people were instructed to respond eitherto shapes with taste words or to taste words with shapes, under conditions that definedthe stimulus–response relations either congruently (e.g., respond round to ‘‘sweet’’) orincongruently (e.g., respond angular to ‘‘sweet’’). Both task and congruencemattered: First,the participants were more accurate and faster when responding to taste words with shapesthan when responding to shapes with taste words. And second, the participants respondedmore rapidly with congruent as compared to with incongruent pairings of stimuli andresponses.

One important question, relating to Experiment 1, is whether the semantic basis oftaste words and shapes may also apply to tastants. Research by Velasco et al. (2015a)demonstrated that the ways in which people match taste words and tastants to shapesfollow similar patterns (as might be expected).2 Sweet tends to associate with round shapeswhilst bitter, sour, salty, and salty associate with angular shapes. One important directionfor future research concerns the evaluation of the semantic space of both taste words andtastants. For example, one could examine the common semantic space for taste words andtastes by running either a semantic differentiation study or a similarity rating experimenton a stimulus set that included both tastes and taste words. Findings of Gallace, Boschin& Spence (2011) with different foods, however, do suggest that foods with specific tastequalities aligned in semantic differential scales (see also Ngo et al., 2013).

Although the present research focused on taste words and shapes, it is worth consideringwhether the results would be similar if we had used shape words instead of shapes.Presumably, shape words would operate semantically like shapes per se, at least to theextent that taste words operate semantically like tastes (although in both cases there maybe some interesting differences between the connotative meanings of perceptual stimuliand the analogous words, as Osgood, 1960, suggested with colors and color words). Insome instances, words may connote ‘prototypes’ that cannot easily be realized in particularstimuli. Such a matter may be an interesting direction for future research.

Nearly two decades ago, Marks (1996) highlighted that ‘‘The correspondences betweenprimary perceptual meanings and secondary linguistic ones need not be perfect—languageand perception do not necessarily carve the world up in precisely the same way (cf. Miller &Johnson-Laird, 1976)—but the connections are nevertheless strong ’’ (p. 49). The results ofExperiment 2 extend previous work on taste/shape associations and taste and shapeinformation processing to taste words and shapes (e.g., Becker et al., 2011; Gal, Wheeler &Shiv, 2007). As noted before, in the English language, for example, the use of shape-relatedwords such as ‘‘sharp’’ to describe tastes has a long history (Marks, 1978; Williams, 1976;Yu, 2003). This said, even though the effects found in Experiment 2 were small, so too wereearlier taste/shape congruence effects reported with perceptual stimuli (Liang et al., 2013;note, however, that comparing effect sizes across experimental paradigms is not an easytask given their different nature), and these effects prove noteworthy given the seeminglyunrelated nature of basic taste words and shapes.

How to interpret the fact that taste word/shape correspondences gave rise to congruenceeffects? In order to answer this question, it is important to highlight the fact that the tasksincluded in Experiment 2 required the participants to learn specific associations (either

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 14/20

congruent or incongruent). This said, it is reasonable to assert that there is an implicitrelation between specific taste words and shape curvature. In other words, participantsmay find it easier to respond to a learned stimulus mapping that has a stronger (implicit)association than to one that has a weaker association, thus responding faster to the former.How is such an implicit relationship built? The results of Experiment 1, together withthose of Velasco et al. (2015a) and Velasco et al. (2016), provide some clues in support ofa hedonic association, and preliminary experimental data for a sensory-discriminativeassociation (see also e.g.,Marks, 1978;Marks, 2013; Parise & Spence, 2013; Spence, 2011, forreviews on possible mechanisms underlying crossmodal correspondences). These ideas areconsistent with an affective account of certain associations across the senses (e.g., Collier,1996; Palmer et al., 2013). Given that intensity and hedonics are also influenced by otherlow-level visual properties and shape aesthetic features (e.g., Palmer, Schloss & Sammartino,2013), which influence taste/shape correspondences (Salgado-Montejo et al., 2015), itshould be reasonable to extend the present results to other visual attributes (e.g., shapesymmetry).

The two experiments reported here were conducted online. Even though there is stilldebate as to the limitations and scope of online research, a number of studies have startedto suggest that, in many situations, online research can mimic laboratory results (not tomention the access to larger and more varied samples of participants; see Woods et al.,2015, for a review of perception research online). Nonetheless, the differences in hardware,software, and contexts across online participants may certainly introduce some error(though the larger samples likely compensate, see also Chetverikov & Upravitelev, in press).

The results of the present study can be of interest to both researchers and practitionersin the context of food and drink design, given that specific stimulus combinations may leadto subtle variations in the processing of taste-related information. Indeed, even thoughresearch will be undoubtedly needed, perhaps, by changing the shape of a packaging, plate,or food, it may be possible to influence, for example, the expected and perceived sweetnessof foods or drinks without touching their actual concentration (Velasco et al., 2014).

To conclude, people appear to respond differently to tastes and shapes when themappings are consistent rather than inconsistent with the correspondence—an additionalpiece of evidence to suggest that taste words and shapes share an abstract semantic networkand that the existence of crossmodally shared locations in semantic space ipso facto defineor characterize crossmodal congruence.

ADDITIONAL INFORMATION AND DECLARATIONS

FundingThe authors received no funding for this work.

Competing InterestsThe authors declare there are no competing interests. Andy T. Woods is an employee ofXperiment, UK.

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 15/20

Author Contributions• Carlos Velasco conceived and designed the experiments, performed the experiments,analyzed the data, wrote the paper, prepared figures and/or tables, reviewed drafts of thepaper.• Andy T. Woods conceived and designed the experiments, performed the experiments,analyzed the data, wrote the paper, reviewed drafts of the paper.• Lawrence E. Marks and Charles Spence conceived and designed the experiments, wrotethe paper, reviewed drafts of the paper.• Adrian David Cheok reviewed drafts of the paper.

Human EthicsThe following information was supplied relating to ethical approvals (i.e., approving bodyand any reference numbers):

Central University Research Ethics Committee at the University of Oxford (MS-IDREC-C1-2014-056).

Data AvailabilityThe following information was supplied regarding data availability:

The raw data is included in the Supplemental Information.

Supplemental InformationSupplemental information for this article can be found online at http://dx.doi.org/10.7717/peerj.1644#supplemental-information.

REFERENCESAlbertazzi L, Canal L, Dadam J, Micciolo R. 2014. The semantics of biological forms.

Perception 43:1365–1376 DOI 10.1068/p7794.BarM, Neta M. 2006.Humans prefer curved visual objects. Psychological Science

17:645–648 DOI 10.1111/j.1467-9280.2006.01759.x.Becker L, Van Rompay TJL, Schifferstein HNJ, GaletzkaM. 2011. Tough package,

strong taste: the influence of packaging design on taste impressions and productevaluations. Food Quality and Preference 22:17–23DOI 10.1016/j.foodqual.2010.06.007.

Cavanaugh LA, MacInnis DJ, Weiss AM. 2015. Perceptual dimensions differentiateemotions. Cognition and Emotion 1–16 DOI 10.1080/02699931.2015.1070119.

Chetverikov A, Upravitelev P. Online versus offline: the Web as a medium for responsetime data collection. Behavior Research Methods. In PressDOI 10.3758/s13428-015-0632-x.

Cliff N. 1996. Answering ordinal questions with ordinal data using ordinal statistics.Multivariate Behavioral Research 31:331–350 DOI 10.1207/s15327906mbr3103_4.

Collier GL. 1996. Affective synesthesia: extracting emotion space from simple perceptualstimuli.Motivation and Emotion 20:1–32 DOI 10.1007/BF02251005.

Cytowic RE. 1993. The man who tasted shapes. New York: G. P. Putnam’s Sons.

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 16/20

Cytowic RE,Wood FB. 1982. Synesthesia: II. Psychophysical relations in the synesthesiaof geometrically shaped taste and colored hearing. Brain and Cognition 1:36–49DOI 10.1016/0278-2626(82)90005-7.

Erceg-Hurn DM,Mirosevich VM. 2008.Modern robust statistical methods: an easyway to maximize the accuracy and power of your research. American Psychologist63:591–601 DOI 10.1037/0003-066X.63.7.591.

Fairhurst MT, Pritchard D, Ospina D, Deroy O. 2015. Bouba-Kiki in the plate: com-bining crossmodal correspondences to change flavour experience. Flavour 4:22DOI 10.1186/s13411-015-0032-2.

Gal D,Wheeler SC, Shiv B. 2007. Cross-modal influences on gustatory perception.Available at SSRN: ssrn.com/abstract=1030197 .

Gallace A, Boschin E, Spence C. 2011. On the taste of Bouba and ‘‘Kiki’’: an explorationof word–food associations in neurologically normal participants. Cognitive Neuro-science 2:34–46 DOI 10.1080/17588928.2010.516820.

Ghoshal T, Boatwright P, Malika M. 2015. Curvature from all angles: an integrativereview and implications for product design. In: Batra R, Seifert C, Brei D, eds. Thepsychology of design: creating consumer appeal . New York: Routledge 91–105.

Guerdoux E, Trouillet R, Brouillet D. 2014. Olfactory–visual congruence effects stableacross ages: yellow is warmer when it is pleasantly lemony. Attention, Perception, &Psychophysics 76:1280–1286 DOI 10.3758/s13414-014-0703-6.

HollandMK,WertheimerM. 1964. Some physiognomic aspects of naming,or, maluma and takete revisited. Perceptual and Motor Skills 19:111–117DOI 10.2466/pms.1964.19.1.111.

Karwoski TF, Odbert HS, Osgood CE. 1942. Studies in synesthetic thinking: II. The roleof form in visual responses to music. The Journal of General Psychology 26:199–222DOI 10.1080/00221309.1942.10545166.

Kaufman L, Rousseeuw PJ. 2005. Finding groups in data: an introduction to clusteranalysis. Hoboken: John Wiley & Sons.

Kenneth JH. 1923.Mental reactions to smell stimuli. Psychological Review 30:77–79DOI 10.1037/h0068405.

KöhlerW. 1929.Gestalt psychology . New York: Liveright.Liang P, Roy S, ChenML, Zhang GH. 2013. Visual influence of shapes and semantic

familiarity on human sweet sensitivity. Behavioural Brain Research 253:42–47DOI 10.1016/j.bbr.2013.07.001.

Lindauer MS. 1990. The effects of the physiognomic stimuli taketa and maluma onthe meanings of neutral stimuli. Bulletin of the Psychonomic Society 28:151–154DOI 10.3758/BF03333991.

Lyman B. 1979. Representation of complex emotional and abstract meanings by simpleforms. Perceptual and Motor Skills 49:839–842 DOI 10.2466/pms.1979.49.3.839.

Marks LE. 1978. The unity of the senses: interrelations among the modalities. New York:Academic Press.

Marks LE. 1996. On perceptual metaphors.Metaphor and Symbol 11:39–66DOI 10.1207/s15327868ms1101_3.

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 17/20

Marks LE. 2013. Weak synaesthesia in perception and language. In: Simner J, HubbardE, eds. The Oxford handbook of synaesthesia. Oxford: Oxford University Press761–789.

Martino G, Marks LE. 1999. Perceptual and linguistic interactions in speededclassification: tests of the semantic coding hypothesis. Perception 28:903–924DOI 10.1068/p2866.

Martino G, Marks LE. 2001. Synesthesia: strong and weak. Current Directions in Psycho-logical Science 10:61–65 DOI 10.1111/1467-8721.00116.

Miller GA, Johnson-Laird PN. 1976. Language and perception. Cambridge: BelknapPress.

NgoMK, Velasco C, Salgado A, Boehm E, O’Neill D, Spence C. 2013. Assessing cross-modal correspondences in exotic fruit juices: the case of shape and sound symbolism.Food Quality and Preference 28:361–369 DOI 10.1016/j.foodqual.2012.10.004.

Noguchi K, Gel YR, Brunner E, Konietschke F. 2012. nparLD: an R Software Package forthe nonparametric analysis of longitudinal data in factorial experiments. Journal ofStatistical Software 50:1–23.

Osgood CE. 1960. The cross-cultural generality of visual-verbal synesthetic tendencies.Behavioral Science 5:146–169 DOI 10.1002/bs.3830050204.

Osgood CE. 1964. Semantic differential technique in the comparative study of cultures.American Anthropologist 66:171–200.

Osgood CE, Suci GJ, Tannenbaum PH. 1957. The measurement of meaning . Urbana:University of Illinois Press.

Palmer SE, Langlois TA, Schloss KB. 2015.Music-to-color associations of single-line piano melodies in non-synesthetes.Multisensory Research 29:157–193DOI 10.1163/22134808-00002486.

Palmer SE, Schloss KB, Sammartino J. 2013. Visual aesthetics and human preference.Annual Review of Psychology 64:77–107 DOI 10.1146/annurev-psych-120710-100504.

Palmer SE, Schloss KB, Xu Z, Prado-León LR. 2013.Music–color associations aremediated by emotion. Proceedings of the National Academy of Sciences of the UnitedStates of America 110:8836–8841 DOI 10.1073/pnas.1212562110.

Parise CV. 2015. Crossmodal correspondences: standing issues and experimentalguidelines.Multisensory Research 29:7–28 DOI 10.1163/22134808-00002502.

Parise CV, Knorre K, Ernst MO. 2014. Natural auditory scene statistics shapes humanspatial hearing. Proceedings of the National Academy of Sciences of the United States ofAmerica 111:6104–6108 DOI 10.1073/pnas.1322705111.

Parise C, Spence C. 2013. Audiovisual cross-modal correspondences in the generalpopulation. In: Simner J, Hubbard E, eds. The Oxford handbook of synaesthesia.Oxford: Oxford University Press 790–815.

Piqueras-Fiszman B, Alcaide J, Roura E, Spence C. 2012. Is it the plate or is it thefood? Assessing the influence of the color (black or white) and shape of the plateon the perception of the food placed on it. Food Quality and Preference 24:205–208DOI 10.1016/j.foodqual.2011.08.011.

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 18/20

R Core Team. 2015. R: a language and environment for statistical computing . Vienna: RFoundation for Statistical Computing. Available at https://www.R-project.org/ .

Ramachandran VS, Hubbard EM. 2001. Synaesthesia–a window into perception,thought and language. Journal of Consciousness Studies 8:3–34.

Salgado-Montejo A, Alvarado J, Velasco C, Salgado CJ, Hasse K, Spence C. 2015.The sweetest thing: the influence of angularity, symmetry, and number of ele-ments on shape-valence and shape-taste matches. Frontiers in Psychology 6:1382DOI 10.3389/fpsyg.2015.01382.

Schifferstein HNJ, Tanudjaja I. 2004. Visualizing fragrances through colors: themediating role of emotions. Perception 33:1249–1266 DOI 10.1068/p5132.

Spence C. 2011. Crossmodal correspondences: a tutorial review. Attention, Perception, &Psychophysics 73:971–995 DOI 10.3758/s13414-010-0073-7.

Spence C, Deroy O. 2014. Tasting shapes: a review of four hypotheses. Theoria et HistoriaScientiarum 10:207–238 DOI 10.12775/ths-2013-0011.

Spence C, NgoMK. 2012. Assessing the shape symbolism of the taste, flavour, andtexture of foods and beverages. Flavour 1:12 DOI 10.1186/2044-7248-1-12.

Stein BE, ed. 2012. The new handbook of multisensory processing . Cambridge: MIT Press.Stewart PC, Goss E. 2013. Plate shape and colour interact to influence taste and quality

judgments. Flavour 2:27 DOI 10.1186/2044-7248-2-27.Velasco C, Salgado-Montejo A, Marmolejo-Ramos F, Spence C. 2014. Predictive

packaging design: tasting shapes, typographies, names, and sounds. Food Quality andPreference 34:88–95 DOI 10.1016/j.foodqual.2013.12.005.

Velasco C,Woods A, Deroy O, Spence C. 2015a.Hedonic mediation of the crossmodalcorrespondence between taste and shape. Food Quality and Preference 41:151–158DOI 10.1016/j.foodqual.2014.11.010.

Velasco C,Woods AT, Hyndman S, Spence C. 2015b. The taste of typeface. i-Perception6:1–10 DOI 10.1068/i0697sas.

Velasco C,Woods AT, Liu J, Spence C. 2016. Assessing the role of taste intensity andhedonics in taste–shape correspondences.Multisensory Research 29:209–221DOI 10.1163/22134808-00002489.

Walker P. 2012. Cross-sensory correspondences and cross talk between dimensions ofconnotative meaning: visual angularity is hard, high-pitched, and bright. Attention,Perception, & Psychophysics 74:1792–1809 DOI 10.3758/s13414-012-0341-9.

Walker L, Walker P. Cross-sensory mapping of feature values in the size-brightness cor-respondence can be more relative than absolute. Journal of Experimental Psychology:Human Perception and Performance. In Press.

Walker L, Walker P, Francis B. 2013. A common scheme for cross-sensory correspon-dences across stimulus dimensions. Perception 41:1186–1192.

Wan X,Woods AT, Van den Bosch J, McKenzie KJ, Velasco C, Spence C. 2014. Cross-cultural differences in crossmodal correspondences between basic tastes and visualfeatures. Frontiers in Psychology 5:1365 DOI 10.3389/fpsyg.2014.01365.

Williams JM. 1976. Synaesthetic adjectives: a possible law of semantic change. Language52:461–478 DOI 10.2307/412571.

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 19/20

Woods AT, Velasco C, Levitan CA,Wan X, Spence C. 2015. Conducting perceptionresearch over the internet: a tutorial review. PeerJ 3:e1058 DOI 10.7717/peerj.1058.

Yu N. 2003. Synesthetic metaphor: a cognitive perspective. Journal of Literary Semantics32:19–34 DOI 10.1515/jlse.2003.001.

Velasco et al. (2016), PeerJ, DOI 10.7717/peerj.1644 20/20